Minsoo Ha
Minsoo Ha (하민수)
B.S. in Mathematics
Pohang University of Science and Technology (POSTECH)
haminsoo0926 [at] gmail [dot] com

I am an undergraduate student at POSTECH, majoring in Mathematics. I am currently a research assistant at the Interactive Machine Learning Lab advised by Prof. Kwang-Sung Jun.

My research interests lie in Machine learning theory, Neuro-symbolic AI, and the algorithmic foundations of large language models. I am especially drawn to problems where mathematical structure can be used to make learning systems more reliable, interpretable, and provably correct.

Alongside my research, I am deeply interested in software engineering and passionate about building products that deliver real-world impact. In line with this commitment, I am currently building a service KnowGem, a knowledge-graph workspace for hallucination-free reasoning over documents.

News

Apr 2026 Joined the Interactive Machine Learning Lab @ POSTECH as an undergraduate research assistant (advisor: Prof. Kwang-Sung Jun).
Sep 2025 Co-founded KnowGem and started research at the ML Lab @ POSTECH on LLM reasoning & RL (advisor: Prof. Sangdon Park).
Jan 2025 Joined LinqAlpha as a Fundamental Research Engineer, working on RAG for finance, Text-to-SQL, and STT workflows.
Dec 2024 Represented POSTECH in Elite League S2 (Coupang Play, a.k.a. University War) — selected as one of four student representatives.
Aug 2024 Joined FBA Quant, applying ML methods to quantitative finance.
Oct 2023 Completed military service (ROK Army, reserve training instructor).
Feb 2020 Began B.S. in Mathematics @ POSTECH.

Research & Projects

KnowGem
A focused workspace that turns scattered documents into a navigable graph context and validates reasoning outputs with Higher-Order Logic. Aims to enable hallucination-free, deductive reasoning grounded in symbolic structure.
project page · React, Express.js, Supabase, AWS, Terraform
Sword RL
End-to-end data-parsing, RL training environment, and automation tool for the KakaoTalk Gamebot game 검키우기. Built on Stable-Baselines3 and Gymnasium.
PPO
A clean, from-scratch PyTorch implementation of PPO compatible with the Gymnasium API, supporting both discrete and continuous control.
code · PyTorch, Gymnasium
DRL Portfolio
A portfolio-optimization system that learns allocation policies with PPO under diverse simulated market regimes, served through a FastAPI training/evaluation UI.
code · PyTorch, Stable-Baselines3, FastAPI

Service

Education